A trust-based detection scheme to explore anomaly prevention in social networks

Abstract

With the increasing popularity of social networks, malicious behaviors in social networks have brought severe security threats. To address the growing problem of malicious activities on social networks, researchers have proposed different detection and mitigation methods. However, the proposed approaches mainly focus on anomaly detection rather than anomaly prevention. In order to stop and reduce security threats in social networks, the anomalous behavior should be detected before it actually happens. Although some work exists, more effective and novel techniques need to be done toward anomaly prevention. In the paper, we explore the issue of anomaly prevention and creatively introduce trust to measure the similarity between users. Finally, a trust-based anomaly detection scheme is proposed, which can find collective anomaly before it brings greater security risks. The main contributions of our paper are: (1) constructing similar trust graph; (2) improving spectral clustering algorithm; (3) detecting collective anomalies in real time. In order to evaluate the effectiveness of our anomaly detection scheme, a series of test scenarios are developed. Three types of collaborative attacks are simulated on OMNeT++ platform. Simulation results show the good performance of the proposed anomaly detection scheme.